Chapel 2.2 update (#46593)

* shorten version number validations per reviewer feedback
* rename set_lib_path per reviewer feedback
* Add E4S tag
* Set CHPL_CUDA_PATH to ensure Chapel installer finds the right package
* Update ROCm dependency for Chapel 2.2
* Fix llvm-amdgpu and CHPL_TARGET_* for llvm=bundled
* Ensure CHPL_TARGET_COMPILER is set to "llvm" when required (llvm=spack
   or +cuda or +rocm).
* Ensure CHPL_TARGET_{CC,CXX} are only set when using llvm=spack or llvm=none
* Use hip.prefix to set CHPL_ROCM_PATH
   Since we might not directly depend on llvm-amdgpu, thus it might
   not appear in our spec
* limit m4 dependency to +gmp
* limit names of env vars created from variants
* Ensure that +cuda and +rocm variants are Sticky

The concretizer should never be permitted to select GPU support, because
it's only meaningful and functional when the appropriate hardware is actually
available, and the concretizer cannot reliably determine that.

Also: Chapel's GPU support includes alot of complicated dependencies
and constraints, so leaving that choice free to the concretizer leads to alot
of extraneous and confusing messages when failing to concretize a
non-GPU-enabled spec.

Co-authored-by: Dan Bonachea <dobonachea@lbl.gov>

1 files changed, 44 insertions, 27 deletions

diff --git a/var/spack/repos/builtin/packages/chapel/package.py b/var/spack/repos/builtin/packages/chapel/package.py
index 773b4366c8..0538221e35 100644
--- a/var/spack/repos/builtin/packages/chapel/package.py
+++ b/var/spack/repos/builtin/packages/chapel/package.py
@@ -36,7 +36,7 @@ class Chapel(AutotoolsPackage, CudaPackage, ROCmPackage):
 
     homepage = "https://chapel-lang.org/"
 
-    url = "https://github.com/chapel-lang/chapel/archive/refs/tags/2.1.0.tar.gz"
+    url = "https://github.com/chapel-lang/chapel/archive/refs/tags/2.2.0.tar.gz"
     git = "https://github.com/chapel-lang/chapel.git"
 
     test_requires_compiler = True
@@ -48,11 +48,14 @@ class Chapel(AutotoolsPackage, CudaPackage, ROCmPackage):
     # TODO: add chapel-project github account
     maintainers("arezaii", "bonachea")
 
+    tags = ["e4s"]
+
     # See https://spdx.org/licenses/ for a list.
     license("Apache-2.0")
 
     version("main", branch="main")
 
+    version("2.2.0", sha256="9b0ce16ed0b1c777647c3aa852d8d8cacea2c3d8a952548a238f07c0f357a0f9")
     version("2.1.0", sha256="8e164d9a9e705e6b816857e84833b0922ce0bde6a36a9f3a29734830aac168ef")
     version("2.0.1", sha256="47e1f3789478ea870bd4ecdf52acbe469d171b89b663309325431f3da7c75008")
     version("2.0.0", sha256="a8cab99fd034c7b7229be8d4626ec95cf02072646fb148c74b4f48c460c6059c")
@@ -435,15 +438,21 @@ class Chapel(AutotoolsPackage, CudaPackage, ROCmPackage):
 
     conflicts("platform=windows")  # Support for windows is through WSL only
 
+    # Ensure GPU support is Sticky: never allow the concretizer to choose this
+    variant("rocm", default=False, sticky=True, description="Enable AMD ROCm GPU support")
+    variant("cuda", default=False, sticky=True, description="Enable Nvidia CUDA GPU support")
+
     conflicts("+rocm", when="+cuda", msg="Chapel must be built with either CUDA or ROCm, not both")
-    conflicts(
-        "+rocm", when="@:1.99.99", msg="ROCm support in spack requires Chapel 2.0.0 or later"
-    )
+    conflicts("+rocm", when="@:1", msg="ROCm support in spack requires Chapel 2.0.0 or later")
     # Chapel restricts the allowable ROCm versions
-    with when("+rocm"):
+    with when("@2:2.1 +rocm"):
         depends_on("hsa-rocr-dev@4:5.4")
         depends_on("hip@4:5.4")
-        depends_on("llvm-amdgpu@4:5.4")
+    with when("@2.2: +rocm"):
+        depends_on("hsa-rocr-dev@4:5.4,6.0:6.2")
+        depends_on("hip@4:5.4,6.0:6.2")
+    depends_on("llvm-amdgpu@4:5.4", when="+rocm llvm=spack")
+    requires("llvm=bundled", when="+rocm ^hip@6.0:6.2", msg="ROCm 6 support requires llvm=bundled")
 
     conflicts(
         "comm_substrate=unset",
@@ -463,7 +472,7 @@ class Chapel(AutotoolsPackage, CudaPackage, ROCmPackage):
 
     conflicts(
         "^python@3.12:",
-        when="@:2.0.99",
+        when="@:2.0",
         msg="Chapel versions prior to 2.1.0 may produce SyntaxWarnings with Python >= 3.12",
     )
 
@@ -508,7 +517,7 @@ class Chapel(AutotoolsPackage, CudaPackage, ROCmPackage):
     # with an error about missing plugin-api.h
     depends_on("binutils+gold+ld+plugins+headers", when="llvm=bundled")
 
-    depends_on("m4")
+    depends_on("m4", when="gmp=bundled")
 
     # Runtime dependencies:
     # Note here "run" is run of the Chapel compiler built by this package,
@@ -566,12 +575,18 @@ class Chapel(AutotoolsPackage, CudaPackage, ROCmPackage):
 
     def setup_chpl_compilers(self, env):
         env.set("CHPL_HOST_COMPILER", self.compiler_map[self.spec.compiler.name])
-        env.set("CHPL_TARGET_COMPILER", self.compiler_map[self.spec.compiler.name])
+        if (
+            self.spec.satisfies("+rocm")
+            or self.spec.satisfies("+cuda")
+            or self.spec.satisfies("llvm=spack")
+        ):
+            env.set("CHPL_TARGET_COMPILER", "llvm")
+        else:
+            env.set("CHPL_TARGET_COMPILER", self.compiler_map[self.spec.compiler.name])
 
         # Undo spack compiler wrappers:
         # the C/C++ compilers must work post-install
-        if self.spec.satisfies("+rocm"):
-            env.set("CHPL_TARGET_COMPILER", "llvm")
+        if self.spec.satisfies("+rocm llvm=spack"):
             env.set(
                 "CHPL_LLVM_CONFIG",
                 join_path(self.spec["llvm-amdgpu"].prefix, "bin", "llvm-config"),
@@ -585,15 +600,16 @@ class Chapel(AutotoolsPackage, CudaPackage, ROCmPackage):
             env.set("CHPL_HOST_CXX", real_cxx)
 
         elif self.spec.satisfies("llvm=spack"):
-            env.set("CHPL_TARGET_COMPILER", "llvm")
             env.set("CHPL_LLVM_CONFIG", join_path(self.spec["llvm"].prefix, "bin", "llvm-config"))
             real_cc = join_path(self.spec["llvm"].prefix, "bin", "clang")
             real_cxx = join_path(self.spec["llvm"].prefix, "bin", "clang++")
         else:
             real_cc = self.compiler.cc
             real_cxx = self.compiler.cxx
-        env.set("CHPL_TARGET_CC", real_cc)
-        env.set("CHPL_TARGET_CXX", real_cxx)
+
+        if self.spec.satisfies("llvm=spack") or self.spec.satisfies("llvm=none"):
+            env.set("CHPL_TARGET_CC", real_cc)
+            env.set("CHPL_TARGET_CXX", real_cxx)
 
     def setup_chpl_comm(self, env, spec):
         env.set("CHPL_COMM", spec.variants["comm"].value)
@@ -609,7 +625,7 @@ class Chapel(AutotoolsPackage, CudaPackage, ROCmPackage):
         symlink(self.spec["gasnet"].prefix.src, dst)
 
     def setup_if_not_unset(self, env, var, value):
-        if value != "unset":
+        if value != "unset" and var in self.chpl_env_vars:
             if value == "spack":
                 value = "system"
             env.set(var, value)
@@ -618,7 +634,7 @@ class Chapel(AutotoolsPackage, CudaPackage, ROCmPackage):
         if not is_system_path(prefix):
             env.prepend_path("CPATH", prefix.include)
 
-    def set_lib_path(self, env, prefix):
+    def update_lib_path(self, env, prefix):
         if not is_system_path(prefix):
             env.prepend_path("LD_LIBRARY_PATH", prefix.lib)
             env.prepend_path("LIBRARY_PATH", prefix.lib)
@@ -629,7 +645,7 @@ class Chapel(AutotoolsPackage, CudaPackage, ROCmPackage):
         # variants that appear unused by Spack typically correspond directly to
         # a CHPL_<variant> variable which will be used by the Chapel build system
         for v in self.spec.variants.keys():
-            self.setup_if_not_unset(env, "CHPL_" + v.upper(), self.spec.variants[v].value)
+            self.setup_if_not_unset(env, "CHPL_" + v.upper(), str(self.spec.variants[v].value))
         self.setup_chpl_compilers(env)
         self.setup_chpl_platform(env)
 
@@ -657,10 +673,10 @@ class Chapel(AutotoolsPackage, CudaPackage, ROCmPackage):
             # TODO: why must we add to CPATH to find gmp.h
             # TODO: why must we add to LIBRARY_PATH to find lgmp
             self.prepend_cpath_include(env, self.spec["gmp"].prefix)
-            self.set_lib_path(env, self.spec["gmp"].prefix)
+            self.update_lib_path(env, self.spec["gmp"].prefix)
 
         if self.spec.variants["hwloc"].value == "spack":
-            self.set_lib_path(env, self.spec["hwloc"].prefix)
+            self.update_lib_path(env, self.spec["hwloc"].prefix)
             # Need this for the test env, where it does not appear automatic:
             env.prepend_path("PKG_CONFIG_PATH", self.spec["libpciaccess"].prefix.lib.pkgconfig)
 
@@ -673,22 +689,23 @@ class Chapel(AutotoolsPackage, CudaPackage, ROCmPackage):
         if self.spec.satisfies("+yaml"):
             self.prepend_cpath_include(env, self.spec["libyaml"].prefix)
             # could not compile test/library/packages/Yaml/writeAndParse.chpl without this
-            self.set_lib_path(env, self.spec["libyaml"].prefix)
+            self.update_lib_path(env, self.spec["libyaml"].prefix)
 
         if self.spec.satisfies("+zmq"):
             self.prepend_cpath_include(env, self.spec["libzmq"].prefix)
             # could not compile test/library/packages/ZMQ/hello.chpl without this
-            self.set_lib_path(env, self.spec["libzmq"].prefix)
+            self.update_lib_path(env, self.spec["libzmq"].prefix)
             env.prepend_path("PKG_CONFIG_PATH", self.spec["libsodium"].prefix.lib.pkgconfig)
 
         if self.spec.satisfies("+curl"):
             self.prepend_cpath_include(env, self.spec["curl"].prefix)
             # could not compile test/library/packages/Curl/check-http.chpl without this
-            self.set_lib_path(env, self.spec["curl"].prefix)
+            self.update_lib_path(env, self.spec["curl"].prefix)
 
         if self.spec.satisfies("+cuda"):
             # TODO: why must we add to LD_LIBRARY_PATH to find libcudart?
             env.prepend_path("LD_LIBRARY_PATH", self.spec["cuda"].prefix.lib64)
+            env.set("CHPL_CUDA_PATH", self.spec["cuda"].prefix)
             env.set("CHPL_LOCALE_MODEL", "gpu")
             env.set("CHPL_GPU", "nvidia")
 
@@ -697,9 +714,9 @@ class Chapel(AutotoolsPackage, CudaPackage, ROCmPackage):
             env.set("CHPL_GPU", "amd")
             env.set("CHPL_GPU_ARCH", self.spec.variants["amdgpu_target"].value[0])
             self.prepend_cpath_include(env, self.spec["hip"].prefix)
-            env.set("CHPL_ROCM_PATH", self.spec["llvm-amdgpu"].prefix)
-            self.set_lib_path(env, self.spec["hip"].prefix)
-            self.set_lib_path(env, self.spec["hsa-rocr-dev"].prefix)
+            env.set("CHPL_ROCM_PATH", self.spec["hip"].prefix)
+            self.update_lib_path(env, self.spec["hip"].prefix)
+            self.update_lib_path(env, self.spec["hsa-rocr-dev"].prefix)
         self.setup_chpl_comm(env, self.spec)
 
     def setup_build_environment(self, env):
@@ -716,7 +733,7 @@ class Chapel(AutotoolsPackage, CudaPackage, ROCmPackage):
     @llnl.util.lang.memoized
     def _output_version_long(self):
         if str(self.spec.version).lower() == "main":
-            return "2.2.0"
+            return "2.3.0"
         spec_vers_str = str(self.spec.version.up_to(3))
         return spec_vers_str
 
@@ -724,7 +741,7 @@ class Chapel(AutotoolsPackage, CudaPackage, ROCmPackage):
     @llnl.util.lang.memoized
     def _output_version_short(self):
         if str(self.spec.version).lower() == "main":
-            return "2.2"
+            return "2.3"
         spec_vers_str = str(self.spec.version.up_to(2))
         return spec_vers_str